The present invention relates to a packet switching system and, more particularly, to a packet switching system which, in sending and receiving packets over ATM (Asynchronous Transfer Mode) lines, switches the packets based on the IP (Internet Protocol) routing protocol.
Recent years have seen a surge in traffic of packet communications based on the IP thanks in part to the introduction of novel applications such as the WWW (World Wide Web). IP packet switching was carried out conventionally on a packet-by-packet basis with reference to a routing table prepared according to the IP routing protocol. The switching was conducted primarily by software, and it was desired to increase the speed of the processing in the face of an ever-swelling volume of traffic.
One way to address the need for high-speed processing has been proposed in the form of a packet switching system by P. Newman, et al. "Flow Labeled IP: A Connectionless Approach ATM", Proc. IEEE Infocom, March 1996, pp. 1-10 and Japanese Patent Laid-open (Kokai) No. Hei 8-125692. The proposed packet switching system combines two processes: packet-by-packet processing based on a routing table, and processing of individual ATM cells using ATM switches.
The above packet switching system has a packet processor called IP controller connected to each input/output line of an ATM switch. When IP packets arrive in the form of assembled ATM cells, the cells are sent to the IP controller for a while via the ATM switch. The IP controller reassembles the cells to a IP packet. The destination of each packet to which the cells are reassembled is decided by referring to a routing table as has been done conventionally. The cells are again assembled from the packet and sent via the ATM switch to the output lines bound for their destinations.
If a specific condition is met (e.g., if a predetermined number of packets having the same part of header have arrived), the IP controller allocates a dedicated VC (Virtual Channel) to the packets that were hitherto multiplexed on the same VC as other IP packets. The IP controller then informs the input interface in the ATM switch of a VPI (Virtual Path Identifier) and a VCI (Virtual Channel Identifier) of the input line for the IP packets in question, as well as an output line number and a VPI and a VCI of the output line.
Thereafter, of the IP packets that have arrived at the input interface, those whose headers are partially the same (this group of packets is called a flow hereunder), are switched in the form of ATM cells by the ATM switch without passing through the IP controller. The switched IP packets are placed onto the output line bound for the destination (this operation is called a cut-through hereunder). This eliminates the need for having to switch each packet by referring to the routing table, whereby high-speed packet switching based on the ATM switch is implemented.
Conventionally, it has been necessary for a plurality of packet switching systems to allocate dedicated VCs therebetween for each flow of packets when carrying out a cut-through. Therefore, where large quantities of packets need to be processed, the resource allocation becomes a bottleneck. The following is a list of problems to be solved in conventional apparatus:
(1) Throughput is limited. PA0 (2) The input interface of each ATM switch needs to have connection information set for each flow. Under size constraints of the table permitted for the input interface, only a limited number of flows is handled by cut-through operations. PA0 (3) Delays in the dedicated VC resource allocation prevent the cut-through from providing any appreciable improvement in performance for packet transfers of short holding times. PA0 (4) Where packets are transferred through a plurality of packet switching systems connected in a multi-stage structure, no significant improvement in performance is attained if none of the switching systems performs a cut-through.
On the other hand, a way to provide an idle VC table for retrieving unused VCIs and set a direct channel for transmitting and receiving ATM cells sharing the same destination by setting VCs corresponding to transmission demand has been proposed by U.S. Pat. No. 5,452,296. This way brings realizing a cut-through action in the inside of the ATM switching system and lessen a switching delay. However, since the direct channel is not set in each switching system without a setting time, there is still the problem to be solved that there is little effect of lessening a total switching delay in the entire switching network.